[ Pobierz całość w formacie PDF ]
.The answer was a homegrown distance-vector routing protocol called IGRP.IGRP wasdesigned to be easy to configure and use; in operation, it minimized operational overheads withoutexacerbating convergence times.It also leveraged many of the features that made RIP so useful.Thiswould enable a relatively graceful migration, because most network administrators were quite familiarwith RIP.Despite retaining many of RIP's better features, IGRP was carefully crafted to avoid the RIP's limitations.The end result was that network administrators found IGRP to be vastly superior to RIP, yet familiarenough to mitigate the migration.http://wwwin.cisco.com/cpress/cc/td/cpress/fund/iprf/ip2910.htm (3 of 23) [02/02/2001 11.38.49] IGRPAn Overview of IGRPIGRP is a distance-vector protocol designed for use on interior gateway routers within an autonomoussystem.As explained in "The Mechanics of Routing Protocols," routers that use distance-vector routingprotocols regularly forward all, or part, of their routing tables to their immediate neighbors.This processrecurs periodically, and recursively throughout the AS until all the nodes in that system agree on itstopology and distances to known destinations.IGRP was given an expanded capability set relative to RIP and other distance-vector protocols.One ofthe more revolutionary new features was the way it calculated distance vectors.Unlike previous distancevector protocols that used a single metric for calculating routes, IGRP has a series of metrics, each with awide range of possible values.Each of these metrics can be hashed against a mathematical value, orweight.This enables network administrators to customize the route calculation algorithm according totheir specific needs.These metrics, and their weights, are used to calculate a single, composite routingmetric.This composite metric is used to mathematically compare potential routes to destinations.Additionally, IGRP could support multipath routing.RIP, and its myriad RIP-like variants, could onlyremember a single route to any given destination.These routing protocols were, inherently, unable toconduct multipath routing.IGRP can remember up to four routes to any given destination.This enables itto support multipath routing in a distance-vector network.The practical implications of multipath routingare two new capabilities:Load balancing across two, three, or four linksAutomatic recovery from a failed linkThese features are examined in more detail in the section titled "Multipath Routing."Despite its tremendous success in the market and internetworks around the world, IGRP remains aproprietary routing protocol.The ubiquity of Cisco routers may make IGRP appear open, but it is not.Noother router manufacturer besides Cisco Systems supports it.Cisco has not published the details ofIGRP's internal mechanisms, nor is it likely to.Therefore, it is not possible to examine IGRP to the sameextent that openly specified routing protocols can be examined.Despite this handicap, it is still possibleto gain an appreciation for IGRP by examining some of its configurable attributes, as well as itsfunctional mechanics.MetricsOne of the areas in which IGRP excels is its high degree of flexibility.This flexibility is afforded via itsrouting metrics.Unlike RIP---which has a single, static metric---IGRP uses six metrics:Hop countPacket size (described as Maximum Transmission Unit---MTU)The link's bandwidthDelayLoadingReliabilityhttp://wwwin.cisco.com/cpress/cc/td/cpress/fund/iprf/ip2910.htm (4 of 23) [02/02/2001 11.38.49] IGRPIGRP nodes share information pertaining to all six metrics during table updates, but not all six metricsare used to calculate routes.In fact, only bandwidth, delay, load, and reliability can be used to calculateroutes.The other two, hop count and MTU, facilitate routing in other ways.The flexibility of IGRP extends beyond a mere increase in the number of metrics.Additional flexibilityis afforded in the range of values that each of these metrics can support.The administrator can defineeach of these metrics.IGRP also allows the administrator to establish the default weights for thesemetrics, which are also factored into route calculation.Hop CountConsistent with its heritage, IGRP supports the incrementing of a hop counter as one means ofdetermining how far away specific destinations are.Each router in a path counts as a single hop.UnlikeRIP, IGRP does not cap this value at 16.The default maximum hop count is 100; this may be increased,however, up to 255.This enables IGRP to be used in substantially larger networks than RIP can support.It is important to note that although IGRP maintains a hop counter, it does not use this value to determineoptimal paths.Instead, hop counting is just a means of protecting against routing loops.Routes with ahop count in excess of the stated maximum are automatically invalidated by IGRP.Routes whose hopcounts exceed this diameter are considered invalid and are not advertised to neighboring nodes via tableupdates.This fundamental difference in the use of hop counts between RIP and IGRP demonstrates just howdifferent these protocols are.RIP is a simple, if not crude, distance-vector protocol.It determinednetwork distances only by counting the number of routers separating a source and destination machine.Rather than focus on the number of routers in its route calculations, IGRP focuses on the attributes of thetransmission facilities that interconnect the routers.Properly configured, IGRP can discriminate betweenthe performance attributes of different types of transmission facilities.MTUMTU identifies the largest-size datagram that an IGRP router will accept.This value is not used tocalculate routes, nor is it factored into IGRP's composite metric.IGRP nodes communicate with otherIGRP nodes the maximum size of the datagram that they can accommodate.Datagrams larger than arouter's MTU size must be split into two or more datagrams that comply with this size limitation.Very large MTUs may result in a performance penalty, because routers must buffer incoming datagramsuntil they can determine where to forward them.The larger the datagrams, the more buffer space must beused [ Pobierz całość w formacie PDF ]